Hallo Lee and Raghunathan,

Terry has showed some examples, where these diodes are necessary.

In the following application they can be omitted:
Suppose, your designed microcontroller circuit has a built-in power supply consisting of small transformer, rectifier, big storage electrolytic, 5V regulator and microcontroller circuitry. If total
decoupling capacitance at input of regulator is bigger than total decoupling capacitance at output, and by design there is no possibility of shorting regulator's input voltage, diodes can be omitted.

In this situation it's assumed, that NO big inductive loads are connected to regulator's output. Smaller inductive loads can be handled, if adequate decoupling capacitors are used at output of regulator. Then, whole energy coming from switched inductive load will be wasted, while charging decoupling capacitors. This must of course be well designed!

There's a way, how regulators can still be damaged, which is often overlooked:
In our example possibility must be included, that circuitry is connected to outer world by some cables. If now protection circuits at the connection points are designed unproperly, output voltage of regulator can rise to some dangerous level. It can even increase so much, that regulator's output voltage rises beyond input voltage!

Suppose, overvoltage at input signal is shunted via two diodes to 0V and +5V of circuit. Think not only about externally connected ones, but also about chip-internal, perhaps parasitic ones! Then, a certain current will be caused to flow into +5V power supply. If duration of overvoltage situation is only short, decoupling capacitors at regulator's output can handle it, or regulator himself can regulate it. But what, if overvoltage situation is lasting longer, and if decoupling capacitors are demanded too much, and finally current pushed into +5V supply is bigger than total load current, which results in a going out of regulator of regulation? Then, supply voltage rises and rises and rises... As consequence, first your digital circuitry is damaged, and afterwords, when output voltage of regulator becomes bigger than input voltage, also your regulator becomes damaged.

Agreed, stated situation above is rather unprobable. But if there's any possibillity for this to occur, you get a problem...

Keep in mind, that there must always flow a minimum current OUT of regulator. This minimum current is specified in datasheet, or it can be logically concluded from reading datasheet. E.g. For LM7805 regulated output voltage is specified for the range of '5mA to 1.5A'. Then, your minimum current is about 5mA!

You might say, that not to limit current flowing through protection circuitry to +5V power supply is stupid, and you always insert some resistor. Ok, but in fast digital systems that's not so easy. Adding resistors introduces low pass filtering and might be unhelpful.
Equally how you solve this problem, always keep in mind, that currents from overvoltages shunted directly to +5V CAN be problematic.

How can we handle this situation? Remove diode reversely connected from regulator's output to ground and insert a transzorb, like 1.5KE6.8A or SMCJ6.0A (smd type). Now, whatever happens at regulator's output, supply voltage can hardly rise to dangerous level!

Existence of protection circuitry at inputs and outputs, shunting overvoltage directly to supply voltage can cause also another danger:
Suppose your mains cable is unplugged and your circuit gets no power. If now there's a signal at input of circuitry, signal voltage even of standard level represents overvoltage condition, and all the signal voltage is permanently shunted to regulator's output. Of course, regulator is out of regulation, because it's input voltage is zero. So, signal shunted to supply voltage causes a rise of regulator's output voltage. It's evident, that this situation should be limited somehow. If the above transzorb is used at least no danger exists for digital circuitry, but indeed for the regulator! So, in this case this diode connected from output to input of regulator will help, also.

If in this example regulator's output voltage rises and rises, it can happen, that your digitally circuitry is erroneously powered and begins to work! If this is a problem, you need additional measures. The easiest way is, to connect a rather low ohmic resistance from regulator's output to 0V, and to limit current shunted to power supply caused by overvoltage to such a low level, that regulator's output voltage cannot rise to more than about 1V.

Concluding, you see, there are so many different situations (don't forget the undiscussed ones!) demanding for these two diodes, that it's wise to take them as standards for all regulator circuits. Only in the simplest designs, they can be omitted without danger.
Also, have an eye on protection circuits and their surprising consequences.

Kai